Knowledge representation is a fundamental concern of AI research and has given rise to the subproblem of how to efficiently represent knowledge in a computer. We consider various aspects of the problem and have identified key concerns which any practical system applicable to large textual databases must deal with 1) No practical knowledge representation system has been developed at this time which allows the conversion of large amounts of text to machine readable form. This would require a solution to the language understanding problem. This is clearly not solved at this time. 2) There are theoretical limits to knowledge representation that must exist based on known limitations of human performance in interpreting natural language. These limits make it unreasonable to expect a machine solution unless we can achieve a performance with a machine superior to that of a human. The next step in investigating the limits of access to textual data in a large database is a study we have proposed to examine the role of subject knowledge in human performance at rating document pairs for relevance. We have already accumulated multiple human judgments made on two different test sets of document pairs. The judges were all people knowledgeable in the field at some level. We are now approximately half way through a study which seeks to understand better the nature of the human judgment process. A number of judges who are literate in English, but who's area of expertise is not related to molecular biology, are now carrying out the judgment task. The results will make a useful comparison to help delineate the contribution of subject knowledge to judgments. The goal is to better understand how humans use knowledge in selecting documents to meet a defined need.